xref: /linux/fs/btrfs/delayed-ref.h (revision 3652117f854819a148ff0fbe4492587d3520b5e5)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Copyright (C) 2008 Oracle.  All rights reserved.
4  */
5 
6 #ifndef BTRFS_DELAYED_REF_H
7 #define BTRFS_DELAYED_REF_H
8 
9 #include <linux/refcount.h>
10 
11 /* these are the possible values of struct btrfs_delayed_ref_node->action */
12 enum btrfs_delayed_ref_action {
13 	/* Add one backref to the tree */
14 	BTRFS_ADD_DELAYED_REF = 1,
15 	/* Delete one backref from the tree */
16 	BTRFS_DROP_DELAYED_REF,
17 	/* Record a full extent allocation */
18 	BTRFS_ADD_DELAYED_EXTENT,
19 	/* Not changing ref count on head ref */
20 	BTRFS_UPDATE_DELAYED_HEAD,
21 } __packed;
22 
23 struct btrfs_delayed_ref_node {
24 	struct rb_node ref_node;
25 	/*
26 	 * If action is BTRFS_ADD_DELAYED_REF, also link this node to
27 	 * ref_head->ref_add_list, then we do not need to iterate the
28 	 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
29 	 */
30 	struct list_head add_list;
31 
32 	/* the starting bytenr of the extent */
33 	u64 bytenr;
34 
35 	/* the size of the extent */
36 	u64 num_bytes;
37 
38 	/* seq number to keep track of insertion order */
39 	u64 seq;
40 
41 	/* ref count on this data structure */
42 	refcount_t refs;
43 
44 	/*
45 	 * how many refs is this entry adding or deleting.  For
46 	 * head refs, this may be a negative number because it is keeping
47 	 * track of the total mods done to the reference count.
48 	 * For individual refs, this will always be a positive number
49 	 *
50 	 * It may be more than one, since it is possible for a single
51 	 * parent to have more than one ref on an extent
52 	 */
53 	int ref_mod;
54 
55 	unsigned int action:8;
56 	unsigned int type:8;
57 };
58 
59 struct btrfs_delayed_extent_op {
60 	struct btrfs_disk_key key;
61 	u8 level;
62 	bool update_key;
63 	bool update_flags;
64 	u64 flags_to_set;
65 };
66 
67 /*
68  * the head refs are used to hold a lock on a given extent, which allows us
69  * to make sure that only one process is running the delayed refs
70  * at a time for a single extent.  They also store the sum of all the
71  * reference count modifications we've queued up.
72  */
73 struct btrfs_delayed_ref_head {
74 	u64 bytenr;
75 	u64 num_bytes;
76 	/*
77 	 * For insertion into struct btrfs_delayed_ref_root::href_root.
78 	 * Keep it in the same cache line as 'bytenr' for more efficient
79 	 * searches in the rbtree.
80 	 */
81 	struct rb_node href_node;
82 	/*
83 	 * the mutex is held while running the refs, and it is also
84 	 * held when checking the sum of reference modifications.
85 	 */
86 	struct mutex mutex;
87 
88 	refcount_t refs;
89 
90 	/* Protects 'ref_tree' and 'ref_add_list'. */
91 	spinlock_t lock;
92 	struct rb_root_cached ref_tree;
93 	/* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
94 	struct list_head ref_add_list;
95 
96 	struct btrfs_delayed_extent_op *extent_op;
97 
98 	/*
99 	 * This is used to track the final ref_mod from all the refs associated
100 	 * with this head ref, this is not adjusted as delayed refs are run,
101 	 * this is meant to track if we need to do the csum accounting or not.
102 	 */
103 	int total_ref_mod;
104 
105 	/*
106 	 * This is the current outstanding mod references for this bytenr.  This
107 	 * is used with lookup_extent_info to get an accurate reference count
108 	 * for a bytenr, so it is adjusted as delayed refs are run so that any
109 	 * on disk reference count + ref_mod is accurate.
110 	 */
111 	int ref_mod;
112 
113 	/*
114 	 * The root that triggered the allocation when must_insert_reserved is
115 	 * set to true.
116 	 */
117 	u64 owning_root;
118 
119 	/*
120 	 * Track reserved bytes when setting must_insert_reserved.  On success
121 	 * or cleanup, we will need to free the reservation.
122 	 */
123 	u64 reserved_bytes;
124 
125 	/*
126 	 * when a new extent is allocated, it is just reserved in memory
127 	 * The actual extent isn't inserted into the extent allocation tree
128 	 * until the delayed ref is processed.  must_insert_reserved is
129 	 * used to flag a delayed ref so the accounting can be updated
130 	 * when a full insert is done.
131 	 *
132 	 * It is possible the extent will be freed before it is ever
133 	 * inserted into the extent allocation tree.  In this case
134 	 * we need to update the in ram accounting to properly reflect
135 	 * the free has happened.
136 	 */
137 	bool must_insert_reserved;
138 
139 	bool is_data;
140 	bool is_system;
141 	bool processing;
142 };
143 
144 struct btrfs_delayed_tree_ref {
145 	struct btrfs_delayed_ref_node node;
146 	u64 root;
147 	u64 parent;
148 	int level;
149 };
150 
151 struct btrfs_delayed_data_ref {
152 	struct btrfs_delayed_ref_node node;
153 	u64 root;
154 	u64 parent;
155 	u64 objectid;
156 	u64 offset;
157 };
158 
159 enum btrfs_delayed_ref_flags {
160 	/* Indicate that we are flushing delayed refs for the commit */
161 	BTRFS_DELAYED_REFS_FLUSHING,
162 };
163 
164 struct btrfs_delayed_ref_root {
165 	/* head ref rbtree */
166 	struct rb_root_cached href_root;
167 
168 	/* dirty extent records */
169 	struct rb_root dirty_extent_root;
170 
171 	/* this spin lock protects the rbtree and the entries inside */
172 	spinlock_t lock;
173 
174 	/* how many delayed ref updates we've queued, used by the
175 	 * throttling code
176 	 */
177 	atomic_t num_entries;
178 
179 	/* total number of head nodes in tree */
180 	unsigned long num_heads;
181 
182 	/* total number of head nodes ready for processing */
183 	unsigned long num_heads_ready;
184 
185 	u64 pending_csums;
186 
187 	unsigned long flags;
188 
189 	u64 run_delayed_start;
190 
191 	/*
192 	 * To make qgroup to skip given root.
193 	 * This is for snapshot, as btrfs_qgroup_inherit() will manually
194 	 * modify counters for snapshot and its source, so we should skip
195 	 * the snapshot in new_root/old_roots or it will get calculated twice
196 	 */
197 	u64 qgroup_to_skip;
198 };
199 
200 enum btrfs_ref_type {
201 	BTRFS_REF_NOT_SET,
202 	BTRFS_REF_DATA,
203 	BTRFS_REF_METADATA,
204 	BTRFS_REF_LAST,
205 } __packed;
206 
207 struct btrfs_data_ref {
208 	/* For EXTENT_DATA_REF */
209 
210 	/* Root which owns this data reference. */
211 	u64 ref_root;
212 
213 	/* Inode which refers to this data extent */
214 	u64 ino;
215 
216 	/*
217 	 * file_offset - extent_offset
218 	 *
219 	 * file_offset is the key.offset of the EXTENT_DATA key.
220 	 * extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
221 	 */
222 	u64 offset;
223 };
224 
225 struct btrfs_tree_ref {
226 	/*
227 	 * Level of this tree block
228 	 *
229 	 * Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
230 	 */
231 	int level;
232 
233 	/*
234 	 * Root which owns this tree block reference.
235 	 *
236 	 * For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
237 	 */
238 	u64 ref_root;
239 
240 	/* For non-skinny metadata, no special member needed */
241 };
242 
243 struct btrfs_ref {
244 	enum btrfs_ref_type type;
245 	enum btrfs_delayed_ref_action action;
246 
247 	/*
248 	 * Whether this extent should go through qgroup record.
249 	 *
250 	 * Normally false, but for certain cases like delayed subtree scan,
251 	 * setting this flag can hugely reduce qgroup overhead.
252 	 */
253 	bool skip_qgroup;
254 
255 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
256 	/* Through which root is this modification. */
257 	u64 real_root;
258 #endif
259 	u64 bytenr;
260 	u64 len;
261 	u64 owning_root;
262 
263 	/* Bytenr of the parent tree block */
264 	u64 parent;
265 	union {
266 		struct btrfs_data_ref data_ref;
267 		struct btrfs_tree_ref tree_ref;
268 	};
269 };
270 
271 extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
272 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
273 extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
274 extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
275 
276 int __init btrfs_delayed_ref_init(void);
277 void __cold btrfs_delayed_ref_exit(void);
278 
279 static inline u64 btrfs_calc_delayed_ref_bytes(const struct btrfs_fs_info *fs_info,
280 					       int num_delayed_refs)
281 {
282 	u64 num_bytes;
283 
284 	num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_delayed_refs);
285 
286 	/*
287 	 * We have to check the mount option here because we could be enabling
288 	 * the free space tree for the first time and don't have the compat_ro
289 	 * option set yet.
290 	 *
291 	 * We need extra reservations if we have the free space tree because
292 	 * we'll have to modify that tree as well.
293 	 */
294 	if (btrfs_test_opt(fs_info, FREE_SPACE_TREE))
295 		num_bytes *= 2;
296 
297 	return num_bytes;
298 }
299 
300 static inline u64 btrfs_calc_delayed_ref_csum_bytes(const struct btrfs_fs_info *fs_info,
301 						    int num_csum_items)
302 {
303 	/*
304 	 * Deleting csum items does not result in new nodes/leaves and does not
305 	 * require changing the free space tree, only the csum tree, so this is
306 	 * all we need.
307 	 */
308 	return btrfs_calc_metadata_size(fs_info, num_csum_items);
309 }
310 
311 static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
312 					  int action, u64 bytenr, u64 len,
313 					  u64 parent, u64 owning_root)
314 {
315 	generic_ref->action = action;
316 	generic_ref->bytenr = bytenr;
317 	generic_ref->len = len;
318 	generic_ref->parent = parent;
319 	generic_ref->owning_root = owning_root;
320 }
321 
322 static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref, int level,
323 				       u64 root, u64 mod_root, bool skip_qgroup)
324 {
325 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
326 	/* If @real_root not set, use @root as fallback */
327 	generic_ref->real_root = mod_root ?: root;
328 #endif
329 	generic_ref->tree_ref.level = level;
330 	generic_ref->tree_ref.ref_root = root;
331 	generic_ref->type = BTRFS_REF_METADATA;
332 	if (skip_qgroup || !(is_fstree(root) &&
333 			     (!mod_root || is_fstree(mod_root))))
334 		generic_ref->skip_qgroup = true;
335 	else
336 		generic_ref->skip_qgroup = false;
337 
338 }
339 
340 static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
341 				u64 ref_root, u64 ino, u64 offset, u64 mod_root,
342 				bool skip_qgroup)
343 {
344 #ifdef CONFIG_BTRFS_FS_REF_VERIFY
345 	/* If @real_root not set, use @root as fallback */
346 	generic_ref->real_root = mod_root ?: ref_root;
347 #endif
348 	generic_ref->data_ref.ref_root = ref_root;
349 	generic_ref->data_ref.ino = ino;
350 	generic_ref->data_ref.offset = offset;
351 	generic_ref->type = BTRFS_REF_DATA;
352 	if (skip_qgroup || !(is_fstree(ref_root) &&
353 			     (!mod_root || is_fstree(mod_root))))
354 		generic_ref->skip_qgroup = true;
355 	else
356 		generic_ref->skip_qgroup = false;
357 }
358 
359 static inline struct btrfs_delayed_extent_op *
360 btrfs_alloc_delayed_extent_op(void)
361 {
362 	return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
363 }
364 
365 static inline void
366 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
367 {
368 	if (op)
369 		kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
370 }
371 
372 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
373 {
374 	if (refcount_dec_and_test(&ref->refs)) {
375 		WARN_ON(!RB_EMPTY_NODE(&ref->ref_node));
376 		switch (ref->type) {
377 		case BTRFS_TREE_BLOCK_REF_KEY:
378 		case BTRFS_SHARED_BLOCK_REF_KEY:
379 			kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
380 			break;
381 		case BTRFS_EXTENT_DATA_REF_KEY:
382 		case BTRFS_SHARED_DATA_REF_KEY:
383 			kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
384 			break;
385 		default:
386 			BUG();
387 		}
388 	}
389 }
390 
391 static inline u64 btrfs_ref_head_to_space_flags(
392 				struct btrfs_delayed_ref_head *head_ref)
393 {
394 	if (head_ref->is_data)
395 		return BTRFS_BLOCK_GROUP_DATA;
396 	else if (head_ref->is_system)
397 		return BTRFS_BLOCK_GROUP_SYSTEM;
398 	return BTRFS_BLOCK_GROUP_METADATA;
399 }
400 
401 static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
402 {
403 	if (refcount_dec_and_test(&head->refs))
404 		kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
405 }
406 
407 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
408 			       struct btrfs_ref *generic_ref,
409 			       struct btrfs_delayed_extent_op *extent_op);
410 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
411 			       struct btrfs_ref *generic_ref,
412 			       u64 reserved);
413 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
414 				u64 bytenr, u64 num_bytes,
415 				struct btrfs_delayed_extent_op *extent_op);
416 void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info,
417 			      struct btrfs_delayed_ref_root *delayed_refs,
418 			      struct btrfs_delayed_ref_head *head);
419 
420 struct btrfs_delayed_ref_head *
421 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
422 			    u64 bytenr);
423 int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
424 			   struct btrfs_delayed_ref_head *head);
425 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
426 {
427 	mutex_unlock(&head->mutex);
428 }
429 void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
430 			   struct btrfs_delayed_ref_head *head);
431 
432 struct btrfs_delayed_ref_head *btrfs_select_ref_head(
433 		struct btrfs_delayed_ref_root *delayed_refs);
434 
435 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
436 
437 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr_refs, int nr_csums);
438 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
439 void btrfs_inc_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
440 void btrfs_dec_delayed_refs_rsv_bg_inserts(struct btrfs_fs_info *fs_info);
441 void btrfs_inc_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
442 void btrfs_dec_delayed_refs_rsv_bg_updates(struct btrfs_fs_info *fs_info);
443 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
444 				  enum btrfs_reserve_flush_enum flush);
445 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
446 				       u64 num_bytes);
447 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
448 
449 /*
450  * helper functions to cast a node into its container
451  */
452 static inline struct btrfs_delayed_tree_ref *
453 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
454 {
455 	return container_of(node, struct btrfs_delayed_tree_ref, node);
456 }
457 
458 static inline struct btrfs_delayed_data_ref *
459 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
460 {
461 	return container_of(node, struct btrfs_delayed_data_ref, node);
462 }
463 
464 #endif
465